feat: Add Python Semantic Kernel sample agent

[Yogeshp-MSFT]

Adds a new Python sample demonstrating how to build a production-ready Microsoft 365 agent using Semantic Kernel with the Microsoft Agent 365 SDK. This is the Python counterpart to the existing .NET Semantic Kernel sample (dotnet/semantic-kernel/).

What's included

New directory: python/semantic-kernel/sample-agent/

File	Purpose
agent.py	Core agent implementation using ChatCompletionAgent with FunctionChoiceBehavior.Auto()
host_agent_server.py	Generic aiohttp hosting server following the shared A365 host pattern
start_with_generic_host.py	Entry point
agent_interface.py	Abstract base class for agent implementations
mcp_tool_registration_service.py	MCP server discovery and Semantic Kernel plugin registration
observability_config.py	A365 observability initialization with token resolver
turn_context_utils.py	TurnContext extraction and BaggageBuilder utilities
token_cache.py	In-memory token caching for auth flows
local_authentication_options.py	Environment-based auth configuration
ToolingManifest.json	MCP server manifest (fallback for SDK discovery)
.env.template	Environment variable reference
pyproject.toml	Project configuration with all dependencies
README.md	Full documentation with Quick Start, architecture, troubleshooting

Features

• Dual LLM support — Azure OpenAI or OpenAI via API key, configurable via USE_AZURE_OPENAI env var
• MCP tool integration — Auto-discovered tools via SDK with ToolingManifest.json fallback; connection caching across turns
• Agentic authentication — Token exchange for Graph API, MCP servers, and observability; Playground mode with AUTH_HANDLER_NAME= (empty)
• Full observability — Nested InvokeAgentScope → InferenceScope → ExecuteToolScope with BaggageBuilder context propagation
• Notification handling — Email and Word comment notification processing
• Conversation continuity — Per-conversation ChatHistory maintained across turns
• Multiple messages + typing indicators — Immediate acknowledgment, background typing loop, final response
• Generic host pattern — Reusable GenericAgentHost compatible with Agents Playground, matching the pattern used by other Python samples
• Proper resource cleanup — Graceful shutdown via app.on_shutdown, MCP plugin close, AI service client close, token cache clear

Dependencies

• semantic-kernel>=1.15.0
• microsoft-agents-hosting-aiohttp>=0.7.0
• microsoft-agents-a365-tooling>=0.1.0
• microsoft-agents-a365-observability-core>=0.1.0
• microsoft-agents-a365-notifications>=0.1.0
• microsoft-agents-a365-runtime>=0.1.0

Full list in pyproject.toml.

Testing

• Verified with Agents Playground (Playground mode, AUTH_HANDLER_NAME= empty)
• Tested MCP tool invocation (mail send via mcp_MailTools)
• Tested both OpenAI and Azure OpenAI LLM paths
• Observability spans visible in console output

Related samples

This follows the same architecture and conventions as:

• python/claude/sample-agent/
• python/openai/sample-agent/
• dotnet/semantic-kernel/sample-agent/ (the .NET equivalent)

---

[Copilot]

Pull request overview

Adds a new Python Semantic Kernel sample agent under python/semantic-kernel/sample-agent/, intended to mirror the existing .NET SK sample and follow the shared Agent 365 “generic host” pattern (MCP tooling, notifications, and observability).

Changes:

• Introduces a Semantic Kernel-based agent (ChatCompletionAgent) with MCP tool discovery/registration and notification handling.
• Adds a reusable aiohttp generic host server with optional JWT auth and notification wiring.
• Provides sample packaging/config/docs (pyproject.toml, .env.template, ToolingManifest.json, README.md) for running in Playground and production modes.

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
python/semantic-kernel/sample-agent/agent.py	Implements the Semantic Kernel agent (LLM config, MCP tools, observability scopes, notifications, cleanup).
python/semantic-kernel/sample-agent/agent_interface.py	Defines the async interface required by the generic host.
python/semantic-kernel/sample-agent/host_agent_server.py	Generic aiohttp host wiring: auth, message handling, typing indicators, notifications, shutdown cleanup.
python/semantic-kernel/sample-agent/mcp_tool_registration_service.py	MCP server discovery via SDK and SK plugin registration, with a manual MCP JSON-RPC fallback.
python/semantic-kernel/sample-agent/observability_config.py	Module-level one-time initialization of the A365 observability SDK.
python/semantic-kernel/sample-agent/turn_context_utils.py	Helpers to extract TurnContext details and build baggage for observability.
python/semantic-kernel/sample-agent/token_cache.py	In-memory cache for agentic tokens used by the observability exporter.
python/semantic-kernel/sample-agent/local_authentication_options.py	Loads local auth-related settings from environment variables.
python/semantic-kernel/sample-agent/start_with_generic_host.py	CLI entrypoint to run the agent with the generic host.
python/semantic-kernel/sample-agent/pyproject.toml	Declares dependencies and build config for the sample.
python/semantic-kernel/sample-agent/ToolingManifest.json	MCP server fallback manifest for tool discovery.
python/semantic-kernel/sample-agent/.env.template	Environment variable template for LLM/auth/tooling/observability configuration.
python/semantic-kernel/sample-agent/README.md	End-to-end sample documentation (setup, architecture, troubleshooting).

[Yogeshp-MSFT]

Hi @biswapm can you review this pr?